Why is CL40 ram faster at rendering video than CL38?

[Joe Bauers]

MSI Raider GE 76 RTX3060 6Gb DDR6 12th Gen i7.  OE Ram is 16gb DDR 5 4800 CL 40.

 

I have confirmed that the new ram (Kingston Fury 4800 CL38) is running with the XMP 3.0 profile in CPU-Z.  Bios settings are not modified in any way. MSI user scenario is in discrete mode set at the extreme performance setting.

 

Using Movavi for video editing and I have tested the following scenarios of a 40 minute video at 4k with 30, 60, and 120 FPS using OE 16GB ram, 64GB CL40 4800 Crucial, and Finally 64GB CL38 4800 Kingston.  I am editing with source and output files on the C drive (2TB Crucial PCIe 4.0 P3 Plus). I tried rendering from C drive as source to my second identical NVME and vice versa but output times are slower by about 5-10% oddly.  

 

Why am I am getting worse rendering times using the CL38 from Kingston than with the CL40 from Crucial?  At the end of the day it doesn't matter if my renders take another 30 seconds or so but I was hoping not to have a decrease in performance when upgrading to what appears to be the fastest ram.

 

If I an going to increase performance, what do you recommend?  Undervolting the CPU? Installing Afterburner or Throttlestop? Modifying bios settings?

 

 

 

 

 

 

 

 

[RONOTHAN##]

There's two possible explanations:

1. The two results are within margin of error of each other. You're talking about a single digit percentage difference between the two results, and odds are the render is not that consistent. You might've had something running in the background causing those 30 second differences. 
2. The default JEDEC of the two kits has different lower timings. There are timings that aren't advertised in the XMP of a memory kit like tRRD_L, tRRD_S, tFAW, and tRFC, and they arguably have a bigger impact on performance than the primaries do. The Kingston kit might run them a bit looser than the Crucial kit does, resulting in the ever so slightly worse performance. Going off your previous post, you've got Hynix based memory on the Kingston kit, and Hynix does tighten up those aforementioned timings a lot tighter than Micron memory found on those Crucial sticks, so if you can edit timings in your BIOS somehow the Kingston kit will be faster. 

[Joe Bauers]

6 minutes ago, RONOTHAN## said:

There's two possible explanations:

1. The two results are within margin of error of each other. You're talking about a single digit percentage difference between the two results, and odds are the render is not that consistent. You might've had something running in the background causing those 30 second differences. 
2. The default JEDEC of the two kits has different lower timings. There are timings that aren't advertised in the XMP of a memory kit like tRRD_L, tRRD_S, tFAW, and tRFC, and they arguably have a bigger impact on performance than the primaries do. The Kingston kit might run them a bit looser than the Crucial kit does, resulting in the ever so slightly worse performance. Going off your previous post, you've got Hynix based memory on the Kingston kit, and Hynix does tighten up those aforementioned timings a lot tighter than Micron memory found on those Crucial sticks, so if you can edit timings in your BIOS somehow the Kingston kit will be faster. 

Interesting. I'm glad that I didn't get faulty ram or the system is doing something abnormal.  If these are normal parameters then that's great, I'm not sure if looser timings are good or bad practice or a sign of higher quality product but if it's working as intended I am relieved.   

 

Either way, those are some advanced settings and it's probably best I don't mess with them in the bios, way over my paygrade!  I think my efforts would be better and more reliably focused on eeking out a little more performance using Throttlestop to undervolt.  Might as well keep the CPU a little cooler when it's rendering.

[RONOTHAN##]

2 minutes ago, Joe Bauers said:

I'm not sure if looser timings are good or bad practice or a sign of higher quality product but if it's working as intended I am relieved.   

Generally the rule of thumb is at a given frequency, tighter (lower) timings means better performance, looser (higher) timings mean more stable. It gets a bit more complicated when you start looking at different clock speeds and memory controller modes, plus there's one timing (tREFI) that likes to be as high as possible. As a stability crutch a lot of DRAM manufacturers will set the SPD of the memory to have those subtimings looser to help improve yields, while Micron tends to set them a bit closer to what they actually are capable of. It's just a difference in mindset from each memory manufacturer, both have their pros and cons, plus a lot of motherboards just ignore those timings anyway so it's not like they make that much of a difference. 

 

8 minutes ago, Joe Bauers said:

I think my efforts would be better and more reliably focused on eeking out a little more performance using Throttlestop to undervolt.  Might as well keep the CPU a little cooler when it's rendering.

Yeah, CPU settings are much more important to overall system performance than 2 or 3 memory timings that might be set 2 or 3 ticks higher, and CPU settings take a lot less time to dial in, especially if you've never done it before. 

[mariushm]

So you're comparing 16 GB oem ram ,  vs 64 GB kits ?

You can get a boost by just the extra capacity, as the memory is less fragmented, and the encoding software can allocate memory much faster.

 

The 64 GB kits may be dual rank sticks , while the oem stick could be single rank stick.

With dual rank sticks, the memory controller could send commands to one rank while using the other rank and therefore the data will be ready in that other rank when the controller is done reading or writing data to the other rank, so it saves a few nanoseconds.

 

 

 

[Bombastinator]

The first one that comes to mind is CAS doesn’t have just one number.  It has a bunch of them.  Only one is advertised in a lot of instances though.  You can lower the 1st numbers by increasing a lot of the others most of the time.  The Cl38 memory may actually be slower en total to the  l40

[porina]

It would be interesting if OP could install Thaiphoon Burner so we can see what the modules actually report themselves as.

 

22 minutes ago, mariushm said:

So you're comparing 16 GB oem ram ,  vs 64 GB kits ?

if OP could clarify, I assume the 16GB was supplied with the laptop. Was it 1x16GB or 2x8GB? Either way, I'd suspect they're 1Rx16 which are not great. If single module that would also lose half potential bandwidth.

 

22 minutes ago, mariushm said:

The 64 GB kits may be dual rank sticks , while the oem stick could be single rank stick.

I'm guessing the original is 1Rx16 and the new ones are 2Rx8, so a good jump in ram performance from that alone. When I did similar on my DDR4 laptop and later desktop testing, I saw tens of % improvement in compute, and less for gaming depending on settings. Rendering (Cinebench, Blender) is an area that doesn't benefit much if at all from ram performance though. 

[Bombastinator]

3 hours ago, porina said:

It would be interesting if OP could install Thaiphoon Burner so we can see what the modules actually report themselves as.

 

if OP could clarify, I assume the 16GB was supplied with the laptop. Was it 1x16GB or 2x8GB? Either way, I'd suspect they're 1Rx16 which are not great. If single module that would also lose half potential bandwidth.

 

I'm guessing the original is 1Rx16 and the new ones are 2Rx8, so a good jump in ram performance from that alone. When I did similar on my DDR4 laptop and later desktop testing, I saw tens of % improvement in compute, and less for gaming depending on settings. Rendering (Cinebench, Blender) is an area that doesn't benefit much if at all from ram performance though. 

There’s just too much B.S. from memory manufacturers these days.  They discovered they can hide how fast a stick actually is by gluing on the heat spreader, and everything went downhill from there.

[Joe Bauers]

 

12 hours ago, porina said:

It would be interesting if OP could install Thaiphoon Burner so we can see what the modules actually report themselves as.

I'll post the Thaiphoon Burner results later today / tomorrow.

 

12 hours ago, porina said:

if OP could clarify, I assume the 16GB was supplied with the laptop. Was it 1x16GB or 2x8GB? Either way, I'd suspect they're 1Rx16 which are not great. If single module that would also lose half potential bandwidth.

2x8GB sticks of Samsung 4800 came as OE.  There were modules only on one side with 4 chips per module:

 

 

[mariushm]

Yeah, you can see on the label, 1Rx16 ... single rank... dual rank is better than single rank if the controller can handle it... and x16 (only 4 chips on stick) is also not good, these chips are often slower than x8 chips (8 chips per rank on stick)

 

[Joe Bauers]

 

58 minutes ago, mariushm said:

Yeah, you can see on the label, 1Rx16 ... single rank... dual rank is better than single rank if the controller can handle it... and x16 (only 4 chips on stick) is also not good, these chips are often slower than x8 chips (8 chips per rank on stick)

 

Awesome, my new ram has 8 chips per stick.